main.py

import gradio as gr
from fastapi import FastAPI
from pydantic import BaseModel
from transformers import T5ForConditionalGeneration, T5Tokenizer
import torch

# Load your fine-tuned model
model_path = "./t5-summarizer"  # Path inside Docker container
model = T5ForConditionalGeneration.from_pretrained(model_path)
tokenizer = T5Tokenizer.from_pretrained(model_path, legacy=False)

device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
model = model.to(device)

app = FastAPI()

class TextInput(BaseModel):
    text: str

@app.post("/summarize/")
def summarize_text(input: TextInput):
    input_text = "summarize: " + input.text.strip().replace("\n", " ")

    inputs = tokenizer.encode(input_text, return_tensors="pt", max_length=512, truncation=True).to(device)
    summary_ids = model.generate(inputs, max_length=150, min_length=30, length_penalty=2.0, num_beams=4, early_stopping=True)

    summary = tokenizer.decode(summary_ids[0], skip_special_tokens=True)
    return {"summary": summary}

# Gradio UI setup
gr.Interface(
    fn=lambda text: summarize_text(TextInput(text=text))["summary"],  # Ensure it returns summary
    inputs=gr.Textbox(label="Input Text"),
    outputs=gr.Textbox(label="Summarized Text"),
    flagging=False  # Disable flagging to prevent permission issues
).launch()